Data retrieval

Historical mouse breeding data was provided by two collaborating facilities. Therefore, this study did not involve any type of animal manipulation, observation, or use. Mouse breeding in the collaborating facilities was performed in line with the UK Animals (Scientific Procedures) Act of 1986.

Mouse breeding data were made available by two collaborating breeding facilities (C1, the Babraham Institute and C2, the Wellcome Sanger Institute). Historical production data were downloaded directly from their breeding management software (MCMS, Mouse Colony Management System, the Wellcome Sanger Institute Data Centre). C1 provided data from January 2014 to October 2018, and C2 provided data from January 2010 to March 2019. The datasets contained information on litter identity, breeding adults’ identities, date of birth, date of death, number of pups born, and number of pups weaned.

Animals, housing, and management

The studied dataset contained a total of 34,949 C57BL/6 litters and 219,975 pups. All mice were housed in trios (two females and one male) in individually ventilated cages (IVC). Details on animals, housing, and management are shown in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Animal, housing, and management characteristics for the collaborating animal facilities.

https://doi.org/10.1371/journal.pone.0236290.t001

Statistical analysis

The historical mouse breeding data that was analyzed represent 58,692 pups in 9,261 litters and 161,283 pups in 25,688 litters from C1 and C2, respectively. Required information was retrieved using Scilab (version 6.0.1, Scilab Enterprises, Rungis, France), resulting in one data line per pup. A total of 11% (C1) and 21% (C2) of the data originally provided was excluded (see S1 Fig), mainly due to incongruent data records, implausibly large litters (more than 13 pups, unless confirmed as correct), unreliable information on number and age of older pups in the cage, or missing information. Male pups at C2 euthanized at day 7 post-partum or later were coded as surviving. Litters with males euthanized before day 7 post-partum were excluded.

Pup mortality before weaning was coded as 0 (survived) or 1 (died) and used as the dependent variable. Environmental and social factors were considered as risk factors for pup death. Independent variables representing environmental factors considered for analysis were Collaborator (C1 or C2), Season (Winter, Spring, Summer, Fall), Month (as an alternative predictor to Season), Weekday, and Year, while independent variables for social risk factors included Dam Age (continuous), Father Age (continuous), Litter Size (number of pups born; continuous), litter Overlap (whether or not older siblings were present at the time of birth of the focus litter; no or yes), Sibling Number (number of older pups in the cage at the time of birth of the focus litter; continuous) and Sibling Age (age of the older siblings; continuous).

The risk of pup death was modelled by mixed logistic regression, using the GLIMMIX procedure in SAS (2018 University Edition, SAS Institute Inc., Cary, NC, USA). Multicollinearity among independent variables was checked by using the variance inflation factor (VIF) and regressing each independent variable on the others. As a consequence, Year, Month, and Father Age were excluded from the analysis.

Data of C1 and C2 were combined into one dataset for analysis. Two separate analyses were run: one on the complete dataset (Model 1) and one on a subset of the data containing only pups born in situations with litter Overlap (Model 2) (see S1 Fig). With the first model (Model 1), we aimed at comparing overlapped with non-overlapped litters, accounting for the effects of the remaining covariates and confounders. With the second model (Model 2), we further evaluated how pup probability of death varied with the variation in the age and number of older siblings, also considering the effects of the remaining covariates and confounders. In both models, litter identity was included as a random effect to account for clustering. The models were built by adding one independent variable of interest (Dam Age, Litter Size, Overlap (Model 1) or Sibling Number and Sibling Age (Model 2) at a time in a stepwise process with bidirectional elimination. Independent variables with p ≤ 0.05 were kept in the model. Weekday, Season, and Collaborator were then tested one at a time as confounders, followed by possible interactions and higher order terms. Pair-wise t-tests were performed to compare least-squares means of Overlap, Weekday, and Season, applying a Tukey-Kramer adjustment for multiple pairwise comparisons.

Outcomes from Model 1: Overlap vs. non-overlapped litters

The estimated probability of pup death was seven (C1) and two (C2) percentage points higher (p < 0.001) in cages with the presence of older siblings compared to cages without an older litter (Fig 1A and 1B). At C1, 31% of the overlapped and 26% of the non-overlapped litters had a total litter loss (all pups dying, Fig 1C), whereas at C2, the corresponding figures were 12% and 6% (Fig 1D). Mean Litter Size at birth was 6.3 ± 2.9 pups with no significant differences between collaborators (p = 0.087).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Probability of pup death and litter mortality distribution.

Probability of pup death in litters without (NO) or with (YES) the presence of older siblings in the cage (litter Overlap) at (A) C1, the Babraham Institute, and (B) C2, the Wellcome Sanger Institute, based on least-square means. Percentage of litters by category of pre-weaning mortality in Overlapped and Non-Overlapped litters at (C) C1 and (D) C2, based on raw data. Numbers within brackets in the x-axis designate lower (left side) and upper (right side) endpoints of mortality range. An open bracket next to a number designates a non-inclusive endpoint.

https://doi.org/10.1371/journal.pone.0236290.g001

Although Weekday and Season were not added to the models as variables of interest, these factors turned out to be confounders to the models, while interacting with Collaborator (p < 0.001, p = 0.010, respectively). The probability of death consistently decreased towards the end of the week at both collaborators, while the effect of Season lacked a consistent pattern between collaborators (available in S2 Fig). C1 was able to provide records on cage cleaning dates per Weekday for the period of 17 months (April 2018 to November 2019) for the colony. Cage change events also peaked in the beginning of the week and were lower towards the end of the week (available in S3 Fig).

Outcomes from Model 2: Effects of Sibling Number, Sibling Age, Dam Age, and Litter Size

The predicted probability for a pup to die as a function of Sibling Number and Age, Dam Age, and Litter Size, is illustrated by Figs 2 and 3. Increased Sibling Number, Sibling Age, and Dam Age were associated (p < 0.001) with an increase in the probability of pups dying at both C1 and C2 (Figs 2 and 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Probability of pup death by Dam Age and Sibling Age.

Predicted probabilities (least-squares means) of a pup to die as a function of Dam Age for three distinct levels of Litter Size (number of pups born), at C1, the Babraham Institute and C2, the Wellcome Sanger Institute. Each line corresponds to predictions for a specific value of Sibling Age, as depicted in the legend next to the top left graph. Predictions were obtained while assuming six older pups in the cage, the most recurrent Weekday (Thursday) and the most common Season (Spring) in the combined dataset.

https://doi.org/10.1371/journal.pone.0236290.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Probability of pup death by Sibling Number and Sibling Age.

Predicted probabilities (least-squares means) of a pup to die as a function of Sibling Number (number of older pups in the cage at the birth of the focal litter) for three distinct levels of Litter Size (number of pups born), at C1, the Babraham Institute and C2, the Wellcome Sanger Institute. Each line corresponds to predictions for a specific value of Sibling Age, as depicted in the legend next to the top left graph. Predictions were obtained while assuming a mean Dam Age of 130d, the most recurrent Weekday (Thursday) and the most common Season (Spring) in the combined dataset.

https://doi.org/10.1371/journal.pone.0236290.g003

Mean litter size at birth for overlapped data was slightly larger (p < 0.001) in C1 (6.1 ± 2.7 pups) compared to C2 (5.8 ± 2.9 pups). Pup death probability seem to be higher in either small or large litters due to the quadratic nature of this relationship (see S2 Table). Prediction estimates (least-square means) indicate that, generally, the risk of pup death increases as Litter Size decreases below four and increases above 11 pups per litter, for all combinations of Sibling Age, Sibling Number and Dam Age.

Pup probability of death

Irrespective of any mortality risks identified, the average litter mortality rates obtained both at C1 (39%) and C2 (14%) were higher than what was previously reported for C57BL/6J mice (8% pre-weaning mortality [17] with pup counting at weaning and 3% mortality at three days post-partum [18] with pup number obtained from video-records). Litter mortality was higher in C1 compared to C2 and no differences were found between collaborators in the number of pups born per litter for the whole dataset. Husbandry differences between C1 and C2 may contribute to the mortality difference between both institutes as cage temperature, as well as nest and bedding amount and quality affect rodents’ breeding performance [11–14]. An additional factor may be the practice of euthanizing male pups in C2. Euthanized males after seven days of age were considered as survivals, but this is an assumption and may have led to an underestimation of mortality. Finally, any differences in accuracy of data entry may have affected the results. C1 had a more consistent and early counting of pups than C2. Thus it is possible that C1 has a better accuracy in detecting the number of pups born compared to C2 where pups born could be underestimated, considering that most of deaths happen within 48h post-partum and dead pups often get cannibalized by the dam, thus not seen by the caretakers. Productivity differences in pup survivability due to the distinct mouse sub-strains between C1 and C2 could also underlie the differences in overall mortality found between C1 and C2.

We found higher mouse pre-weaning risk of death in trios with overlapped litters, i.e. litters born when an older litter was present, compared to non-overlapped litters, accounting for the variation in litter size, dam age, and confounders (weekday and season of birth), which is in agreement with previous experimental findings. In outbred mice derived from the C57BL/6J, BALB/c, and DBA/1J strains, Schmidt et al. [19] found pup mortality to increase with increasing age gap between the litters sharing a cage at a specific time and attributed this to infanticide. Understanding why being born into a cage with an older litter is so dangerous requires information about events around pup death and the condition of dead and dying pups. Pup mortality is often associated with infanticide [16, 20], assuming that cannibalized pups were killed before they were eaten. However, our previous behavior studies suggest that infanticide precedes less than 15% of the cannibalism events [6, 21] and that pups die primarily from other causes than direct killing. Litter asynchrony, which often leads to overlap, is likely to increase unequal competition for access to milk and parental care, trauma caused by trampling and stepping of newborns by the adults or the older siblings, and problems related with increased cage stocking density.

Early access to milk is essential for the survival of newborn pups. Measurements of pup energy losses due to metabolism between nursing bouts, extrapolated for a period of 24 hours, revealed that if pups did not receive milk during their first day of life, they would lose approximately 8% of their birth weight [22], which would likely reduce their chances of survival. The presence of older and consequently heavier, more developed and more mobile pups in the cage may have interfered with the access of the newborns to milk in general and specifically to steal the iron-rich milk that is present in higher concentrations during the first week of lactation [22]. Also, older and heavier siblings may be able to displace light newborn pups more easily from the dam's nipples, as compared to younger siblings. This could partly explain the interaction found in this work between the number and the age of older siblings in the cage, affecting pup death probability. Our results showed that the smaller the age gap between the newborn and the older litter, the lower was the pup probability of death. Thus, older pups which are less than a week older than the younger siblings may not be able to interfere with the youngest access to appropriate amount and quality of nutrition.

The presence of two litters in the cage has been demonstrated to increase parturition duration and affect parental behavior. Adults in trio cages with two litters were observed to care for their newborn pups a total of 20% less time (all the three adults together) than adults with one single litter in the cage [6], while parental investment is known to improve the chances of survival of young mice. In fact, C57BL/6 females which lost their litters entirely have been found to spend more time outside the nest and invested less time in building the nest prior to parturition [23], while the presence of males in cages with breeding females (CD-1) has been demonstrated to increase pup survival by facilitating maternal behavior [24]. Thus, reduced parental care in cages with more than one litter can be one of the mechanisms through which pup survivability is reduced in the presence of an older litter.

Most often, when there are two females sharing a cage, they also share the same nest, and younger lighter pups get clustered together with the older, heavier, and more mobile pups. Data on post-mortem inspection performed in 324 C57BL/6J pups found dead, by the authors of this study, revealed that 24% of the pups had some kind of traumatic lesion, including bite wounds and bruises [25].

Alloparenting, meaning that an individual contributes to parenting young which are not their offspring, is known to take place in both house mice [26] and laboratory mice [6, 16]. House mouse females are able to switch strategy during their reproductive lifetime: Ferrari et al. [27] found that half of the females switched and that young and inexperienced females were more prone to join another female, while older and more experienced females (and, presumably, with a larger body mass and more capable of producing larger amounts of milk) tended to raise their litters solitarily.

The reproductive consequences of communal breeding are complex. In their study of wild house mice populations, Ferrari et al. [27] found an increase in pup mortality (between days 1–3 and 13 days old of the older litter), aggravated with each additional litter that was found together with the focal litter, but another study of the same population found an increase in pup survival in litters reared communally [28]. In contrast, comparing pairs (where alloparenting is impossible) and trios (where alloparenting is common), Garner et al. [16] found no difference in mortality. However, Garner et al. [16] found number of pups weaned per female to be reduced in trios, whereas Wasson [15] found no difference between pairs and trios. The reduction in breeding performance found by Garner et al. [16] was attributed to a possible reproduction suppression in trios before birth. In the present work, however, the differences in breeding performance between overlapped and non-overlapped litters are likely not due to reproductive suppression before birth, because our models accounted for the variation in number of pups born per litter.

Higher stocking density leads to increased humidity and gas concentration in the air, with ammonia levels increasing in 30 ppm with litter growth from birth to weaning [29] and reaching over 100 ppm in individually ventilated cages with trios 4d post-partum with a maximum concentration of over 600 ppm at weaning [30]. The presence of a second litter in the cage substantially increased ammonia levels, which indicated an effect of housing density on ammonia concentration [30]. Whereas the impact of these ammonia levels on newborn mice has not been studied, ammonia levels from 25 to over 600 ppm have been demonstrated to increase pathological scores, destroy the surface layers of the trachea epithelium lining and increase the severity of rhinitis, otitis, tracheitis, and pneumonia in rats and mice [29–32]. The gas concentration problem may be aggravated by the fact that animal care-takers generally tend to avoid cleaning cages when litters were just born (to avoid pup disturbance).

Dam Age

Pup death probability increased as Dam Age increased in both collaborators. In this study, Dam Age and parity were confounded. Thus, it was not possible to distinguish effects on pup death probability of the dam's age and its birthing experience. Decreased productivity and increased mortality in first-parity litters have been reported for a few different species [33, 34], but for mice this subject remains controversial. While first-parity BALB/c and 129/Sv dams were reported to wean fewer pups per litter compared to later parity ones [35], we previously found an increase in pup survival with lower parities [6] in an experimental study conducted in C1, whereas another study did not find any significant differences in pup loss between first- and later parity C57BL/6 or BALB/c dams [5].

The results for Dam Age are in agreement with those from Tarín et al. [10], who found increased pup mortality and incidence of litters with at least one cannibalized pup with increased parity. Tarín et al. [10] also compared breeding performance between dams (F1 of C57BL/6JIco × CBA/JIco) who started their reproductive life at age 70d (young) and 357d (old). The authors found no differences between dam age group on pre-weaning mortality and litter size both at birth and at weaning, but reported that young dams produced F2 litters with higher expectation of survival and body weight than those of old dams.

The effect of maternal age on mortality may be related to the age-related hearing loss that is common to females of the strain studied [36]. Pup vocalizations are important signals of cold stress and distress situations and affect maternal behavior such as pup retrieval and maternal care [37, 38]. Thus, if older C57BL/6 females are not able to hear and communicate with their pups appropriately, pup chances of survival could be reduced through a reduction in maternal care, especially knowing how important maternal care is for pup survival and how delicate pups are when they are born (as reviewed by Latham and Mason [39]).

Number of pups born

The reduced survivability in small litters is in agreement with previous reports for C57BL/6 [6] and F1 hybrid (C57BL/6JIco × CBA/JIco) mice [10]. One possible explanation for this is that a reduced number of pups may reduce the huddle formed by the pups in nest, which is an important mechanism through which newborn rodents maintain their body heat [40].

Litter size may also affect parental care, which is essential for pup survival. Ehret and Bernecker [41] demonstrated that early pup vocalization, which gradually increases in frequency after birth, is essential to maintain maternal attention at high levels, which leads to improved pup weight gain, as compared to pups from dams which were unable to hear them. Therefore, it is possible that a small newly born litter does not emit sufficient vocal cues to ensure sufficient maternal care. Rat litters [42] with one single pup were found to perform only about 10% and 5% of the suckling stimuli performed by litters of 10 and 22 pups. As a consequence, the milk yield of dams (estimated based on adjusted measures of the pups' daily weight gain) raising single pups was only -0.4% to 7.0% of those raising 10 pups, which led one-pup litters to have the lowest growth rate. More than half of the one-pup litters did not show any weight gain in the first five days post-partum. From an evolutionary perspective, a small litter is less worth investing in than a larger litter: Maestripieri and Alleva [43] demonstrated that CD-1 dams of large litters (eight pups) spent more than twice as much time displaying litter defense behaviors against intruder males than dams of small litters (four pups). The increase in pup death probability found in litters of 12 pups and above, on the other hand, may be a result of increased sibling competition for access to milk, as discussed above, and also may represent a ceiling in milk production capacity by the dams [22, 44, 45].

Weekday and Season

In both collaborators, there seemed to be a decrease the probability of pup death towards the end of the week, possibly associated with the timing of cage changes, a management routine which affects the mice as well as the accuracy of mortality detection. In C1, which provided records on cage cleaning dates, these closely mirrored the pattern of pup death probability. To reliably count the number of pups, the cage must be opened and animals moved, something that often only happens at cage cleaning when manipulation is unavoidable. Mortality is, therefore, likely to be more accurately detected for litters born on cage changing days. For example, an eight-pup litter born on a Tuesday with cage cleaning schedule for the same day will be recorded as an eight-pup litter. If two of these pups die in the following 24 hours, this litter's pre-weaning mortality will be recorded as being 25% at weaning. A similar litter born on a Saturday with two pups dying on Sunday, and subsequently cannibalized, will be recorded at the Tuesday cage change as a litter with six pups born with no pre-weaning deaths.

Still, the mouse disturbance hypothesis cannot be disregarded. If pup mortality is affected by cage change, the same pattern would be expected in cage change frequency as in pup death probability per Weekday (of birth). Reeb-Whitaker et al. [1] found a higher pup mortality in cages with weekly changes than those changed once every two weeks. Cage change requires that mice are moved from the dirty to a clean cage, an event that triggers a stress response evidenced by increases in serum corticosterone [46] and general activity [47]. It is possible, therefore, that cage change interferes with parental behavior in breeding cages, which could aggravate pup mortality around those days.